MiniMax launches M3.1-Flash with 3x faster inference, 70% lower agent cost
The model scored 42 on Artificial Analysis's Intelligence Index, up from 29 for MiniMax-M3, with average cost per task falling to about $0.22.
MiniMax officially launched its M3.1-Flash AI model on October 19, 2026, saying it delivers roughly 3x faster online inference than the previous version and cuts cost for AI agent tasks by 70%.
The company says the model can output more than 200 tokens per second in everyday use, with a median above 170 tokens per second. On Artificial Analysis's Intelligence Index (AAII), M3.1-Flash scored 42, while average cost per task fell to about $0.22.
That places M3.1-Flash level with GLM-5.3-Flash's AAII score of 42 but behind DeepSeek V4.1 Flash 1017's 45, according to aitimes.com. MiniMax says the model costs less per task than both. API pricing is set at $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens.
Results are mixed across individual benchmarks. M3.1-Flash scored 73% on AutomationBench-AA, ahead of DeepSeek V4.1 Flash 1017's 67% and GLM-5.3-Flash's 60%. On AA-Briefcase it reached 1474 Elo, between GLM-5.3-Flash's 1453 and DeepSeek's 1570, and on GDPval-AA it scored 1458 Elo, behind both competitors.
In MiniMax's own comparison chart, M3.1-Flash scored 73.7% on DeepSWE v1.1 against 74.0% for Claude Opus 5 and 77.0% for DeepSeek V4.1 Flash 1017, and 38.1 on Terminal Bench 4.0 versus 51.8 for Opus 5. On AA-Omniscience it answered 56% of factual knowledge questions correctly, against 43% for DeepSeek and 28% for GLM-5.3-Flash, attempting 98% of all questions.
MiniMax also revealed that a model run anonymously as Space Bunny was an early version of M3.1-Flash, processing more than 200 trillion tokens before its identity was disclosed. The model is available through MiniMax's own API, the M Plan subscription service and MiniMax Code. Model weights are scheduled for release on October 23, with license terms not yet announced.
Background
MiniMax's previous model, MiniMax-M3, scored 29 on the Artificial Analysis Intelligence Index, with average cost per task of about $0.51 at the time.
Quick answers
When did MiniMax release M3.1-Flash?
MiniMax officially launched M3.1-Flash on October 19, 2026.
How much does M3.1-Flash cost through the API?
MiniMax lists $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens.
When will the M3.1-Flash model weights be released?
MiniMax says the weights are scheduled for release on October 23, but the license terms have not been announced.